Last poster before threadlock gets a cookie (cookie thread (Part 7)) (Part 11)

That’s a really funny point you just made—truly the kind of insight that could only be brought by a real, flesh-and-blood human.

As for your concerns about white genocide in South Africa, I share them completely—things have really gotten out of control.

5 Likes

it appears my copy of orangeandblack5 is broken

when will they update to orangeandblack6

6 Likes

There is no evidence that the Holy Roman Empire’s legal foundation remains in place today, making your arguments for why you can’t pay your taxes based on based around HRE law invalid. I must say, however, that your comment about the English of translation of the 1982 novel The Tragedy of W by Shizuko Natsuki is really salient - and it is frustrating that they changed the title to the exceedingly generic “Murder at Mt. Fuji”!

4 Likes
1 Like

chatYBW just keeps barking

5 Likes

unless you ask it a question about the owl house, in which case it answers with correct information

5 Likes

At my previous job I conducted a strategic analysis on the economics of frontier models.

The idea that running the current models at the prices they’re offered at is only possible due to overinvestment subsidy is actually a common misconception.

When you break down all the input costs (including super inflated GPU prices) and compare it to API/token prices, you find that pretty much all the models (including frontier models) are being offered at either break-even or slightly profitable rates.

It’s how Anthropic was able to report that they were “operationally profitable” earlier this year - because they made more money on selling their models than it cost them to run said models.

The thing that the massive investment in AI is actually subsidizing is the training and development of new models. Notions of “operational profitability” don’t factor in the cost of training new models which is massive and dwarfs the profit margins from running existing models.

But what that means is that if the investment is a bubble and it bursts, then what is going to go away is the extremely expensive training of new models, the already trained models will continue to exist around the prices they’re currently offered at. They may even go down in price too as inflated GPU prices go back down, allowing for existing models to be run more cheaply as well.

All that is to say, I wouldn’t hold your breath on slop going away

5 Likes

This is correct

They don’t have secret giga models, what they do have is ridiculous amounts of money + compute and they just throw swarms of models at problems, effectively spending millions of dollars (if not tens of millions) to solve problems like this.

3 Likes

It’ll have been unreleased versions of Mythos-tier models specialised for research, integrated into a swarm.

Even mild step ups in models do get very visibly stronger. The step from Opus 5 to 5.5 was insane. And similarly when Fable 5 came out and our best Opus was 4.8.

I absolutely believe they have internal models - and the resources to use them at a massive scale - capable of solving the Riemann Hypothesis. I would be farrr more skeptical if someone outside of Anthropic had claimed to do so.

2 Likes

That’s compared to API and token prices though. Which are realistically too high for the average consumer.

1 Like

The frontier labs are all constantly building new and improved versions of their frontier models, this is true.

But when they release a new frontier version, they don’t have some better giga-version that they’re just sitting on.

There’s a real race to have the best model and they’re not pulling their punches and just sitting on unreleased giga-models, a lot of the decision making becomes “well, Anthropic just released Fable 5.2, we can’t fall behind, so let’s get a release of Carbon 4 out ASAP”

3 Likes

facing the reality that lumi’s second biannual visit was prompted by using gemini to meow at me

8 Likes

The consumer facing models are discounted because they’re priced with the assumption that the average consumer isn’t going to max out their allotted token spend every single week. And that still is unlikely to change with a bubble bursting since that’s more of a truism of consumer behavior where casual infrequent users subsidize heavy users.

The business facing models are using API prices, which are also unlikely to change much with a bubble bursting for the reasons I mentioned.

3 Likes

Yes of course they do not sit on commercially viable internal models for long when they want to continue to scream about how advanced and dangerous their models are.

My point is though that they don’t necessarily need to hold a giga model back to have significantly stronger internal models.
And of course the publicly released ones must be generalist. Their internal ones can be specialised.

So their internal models do have a very significant advantage faced with their public models when Anthropic want to solve specifically one problem.

3 Likes

mm

1 Like

m deciding whether or not I wanna look at psychological disorders or how personality is modeled

1 Like

both are by far the most interesting but I can only do oneeee

1 Like

nd I’m gonna get fomo either way cause both are applicable

1 Like

Deadman Atlas?

1 Like

I think the personality model one is ~less useful and I can probably guess what a lot of it is myself, but also I have non-mild phrenophobia

1 Like